Continuity for agents
Recall project decisions and constraints across supported coding tools instead of rebuilding context at every session.
Hyphae Local data and agent memory
Hyphae combines SQL, native structures, lexical and vector search, transactions, and recovery in one Rust engine. Agent Memory turns that foundation into shared context for coding agents.
Keep useful context across sessions while retaining control of the data and its record.
01 / Why it matters
Recall project decisions and constraints across supported coding tools instead of rebuilding context at every session.
Relational data, structures, and search share the same durable storage and transaction authority.
Run embedded or through local clients, with no required cloud account or external database.
Retain backups, recovery records, and proofs that can be checked against the documented snapshot.
02 / How it works
The main stages make the project’s boundaries visible.
CLI, embedded Rust, typed clients, local protocol, and the memory MCP profile
SQL, native structures, and lexical/vector search share transaction boundaries
Write-ahead log, snapshots, lifecycle state, checkpoints, and recovery
Project-scoped context with optional proof material and offline verification
03 / Capabilities
Store, recall, forget, and separate personal, work, and journal context with project boundaries and expiry.
Versioned, bounded queries and native data structures, expanded in the 3.0.0 release.
Lexical, exact-vector, approximate, filtered, and hybrid retrieval within the shared engine.
Embedded or local multi-client use, optional HTTP, typed clients, backup, restore, and diagnostics.
04 / Evidence
The release has public package receipts, a dedicated-hardware report, and an abstract commit-protocol model.
Hardware observations apply to the named workloads and source. The formal model is not a proof of the implementation; recall proofs do not establish the truth of remembered text.
Hyphae